UC Berkeley

The evaluation and design of wearable robotic systems remain significant challenges, particularly in real-world environments where traditional laboratory-based tools are impractical. Existing evaluation methods rely heavily on methods such as surface electromyography (sEMG) and motion capture, which are difficult to deploy continuously in the real workplace. This dissertation introduces ExoMetrix, a unified framework for embedded, continuous evaluation of human-exoskeleton interaction. By integrating onboard sensing, real-time data acquisition, cloud-based processing, and physics-based modeling, ExoMetrix enables estimation of internal biomechanical quantities—including joint torques, spinal compression and shear forces, and muscle loading—without reliance on external measurement systems. This approach transforms exoskeleton evaluation from an episodic, laboratory-bound process into a scalable, real-world capability. Using this framework, this dissertation investigates how different exoskeleton architectures influence internal biomechanical loading. A comparative analysis of a rigid-body exoskeleton and a soft exosuit, with both delivering equivalent assistive extension torques at the hip, reveals that internal spinal loading differs significantly between the two systems. These findings challenge the validity of existing ergonomic models that assume equivalent external torques correspond to equivalent physiological outcomes, particularly with respect to spinal fatigue risk. The results demonstrate that such assumptions do not generalize across device types, highlighting the importance of accounting for device-specific mechanics and human-device interaction dynamics. Despite growing interest in upper-limb assistance, most shoulder exoskeletons remain passive, while active systems are often constrained by weight, cost, and mechanical complexity, limiting their practicality for widespread use. This dissertation addresses these limitations through the design and validation of a novel powered shoulder exoskeleton that demonstrates the feasibility of a single-actuator architecture for delivering independent, variable assistance to both shoulders. Experimental results demonstrate that the device provides torque profiles consistent with a gravity compensation control scheme while adapting to user intent, allowing natural, unconstrained motion across the full range of motion of the arm. Together, these contributions establish a cohesive approach to exoskeleton development that integrates design, control, and embedded evaluation, advancing the deployment of wearable robotic systems in real-world settings.